Below are a few personal projects and AI hacks I’ve built and shipped that I’m proud of:

1. HuberMind AI

A neuroscience-focused AI tool inspired by Dr. Andrew Huberman

🧠 Problem Solved: While I was in my 4th year, I was diagnosed with a brain tumor and had to undergo surgery. During recovery, I got interested in neuroscience and wanted to speed up healing using insights from it. However, watching Dr. Andrew Huberman's 1–2 hour podcasts felt stressful. So, I developed HuberMind AI — a RAG-based tool inspired by Dr. Huberman, where you can just 'Ask Andrew' anything about neuroscience and get intelligent, AI-powered responses.
 

⚙️ Tech Stack: LangChain, OpenAI GPT-4, ChromaDB, FastAPI, Streamlit

🔗 Links: GitHub • Demo Video

🛠️ Challenges & Solutions:
         Challenge: Making answers verifiable and source-linked

         Solution: Integrated RAG pipeline with chunk-based retrieval and source citations

📊 Metrics & Outcomes: Built for learning and submitted as part of a challenge hosted by the Kerala-based community GTech MuLearn. I was one of the selected participants for their AI Bootcamp.

2. SQL Ease

Converts natural language to SQL for non-technical users

🧠 Problem Solved: Manual SQL query writing is a barrier for many analysts. This tool allows NL → SQL generation.

⚙️ Tech Stack: GPT-3.5 Turbo, PostgreSQL, Flask, Tailwind

🔗 Links: SQL Ease GitHub • Live Demo

🛠️ Challenges & Solutions:
             Ambiguous queries → Resolved using prompt rephrasing with examples and few-shot context
 

📊 Metrics & Outcomes: Created during a 6-week virtual accelerator called "Nights and Weekends". It was my first full-stack production-ready web app that received real user feedback and iterative improvements.
 

3. 20-20

CNN-based image classifier for Malayalam film actors

🧠 Problem Solved: Malayalam film fans often debate over Mammootty vs. Mohanlal. This tool classifies actor images using deep learning.

⚙️ Tech Stack: Python, TensorFlow, Flask, HTML, CSS

🔗 Links: GitHub • Live Demo • Loom Walkthrough

🛠️ Challenges & Solutions:
                Limited dataset → Scraped more data using Selenium and Augmented some data and tuned CNN architecture for better accuracy.

📊 Metrics & Outcomes: Achieved 80%+ classification accuracy.
 

4. Pre-Owned

ML model to predict used car prices in India

🧠 Problem Solved: Car buyers and sellers struggle to determine fair pricing. This model predicts resale prices based on features like age, fuel type, kilometers driven, etc.

⚙️ Tech Stack: Python, Flask, HTML, CSS

🔗 Links: GitHub • Live Demo
 

🛠️ Challenges & Solutions:
                At the time, I had no prior knowledge of machine learning, so I used this project as a hands-on opportunity to learn ML fundamentals while building something practical.

📊 Metrics & Outcomes: I was able to learn most of the ML concepts.
 

I’m deeply passionate about AI, fast-paced startups, and love building tools that users actually use. Would love to chat and share more!

Best regards,
Arjun M S
https://fueler.io/arjun_ms
GitHub • LinkedIn • WhatsApp